Security researcher spends $700 in tokens testing AI exploit discovery, unnerved by results
matthew_d_green · x · 2026-11-08
Infosec researcher Matthew Green reports he spent about $700 in tokens letting models throw themselves at many binaries with minimal human intervention, and is 'super uncomfortable' with the results. He's in a dispute with those who claim models can't develop or deploy novel exploits — his experiment suggests otherwise, with real implications for both attackers and defenders.
More from Safety
- JPMorgan CISO: the agent era may flip decades of security toward "private by default" — neal_lathia · 2026-11-09
- Data center logic flips: Argentina's Patagonia courts compute with stranded energy — matt_slotnick · 2026-11-09
- Critical unauthenticated RCE in React Server Components, patched in 19.x point releases — DanielLockyer · 2026-11-09
- 46 years in the making: how a '*' in crypt() led to OpenSSH auth bypass (*sshole) — OwariDa · 2026-11-09
- Polymarket gives Chinese labs just 7% chance of a top AI model by year-end — Polymarket · 2026-11-09
- Security pro slams stale 'patch and unplug' advice in the era of AI cyber superweapons — HankYeomans · 2026-11-09